X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/4440f049a94bd3570271f8f0b2461d7f3f3c4582..41a79db76ab767e22333121194c45306c9c7c6a3:/services/nodemanager/arvnodeman/nodelist.py

diff --git a/services/nodemanager/arvnodeman/nodelist.py b/services/nodemanager/arvnodeman/nodelist.py
index 6d436bee52..0abb3b3a37 100644
--- a/services/nodemanager/arvnodeman/nodelist.py
+++ b/services/nodemanager/arvnodeman/nodelist.py
@@ -1,8 +1,11 @@
 #!/usr/bin/env python
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
 
 from __future__ import absolute_import, print_function
 
-import subprocess
+import subprocess32 as subprocess
 
 from . import clientactor
 from . import config
@@ -12,8 +15,9 @@ import arvados.util
 class ArvadosNodeListMonitorActor(clientactor.RemotePollLoopActor):
     """Actor to poll the Arvados node list.
 
-    This actor regularly polls the list of Arvados node records, and
-    sends it to subscribers.
+    This actor regularly polls the list of Arvados node records,
+    augments it with the latest SLURM node info (`sinfo`), and sends
+    it to subscribers.
     """
 
     def is_common_error(self, exception):
@@ -23,25 +27,35 @@ class ArvadosNodeListMonitorActor(clientactor.RemotePollLoopActor):
         return node['uuid']
 
     def _send_request(self):
-        nodelist = arvados.util.list_all(self._client.nodes)
+        nodelist = arvados.util.list_all(self._client.nodes().list)
 
         # node hostname, state
-        sinfo_out = subprocess.check_output(["sinfo", "--noheader", "--format=%n %t"])
+        sinfo_out = subprocess.check_output(["sinfo", "--noheader", "--format=%n|%t|%f"])
         nodestates = {}
+        nodefeatures = {}
         for out in sinfo_out.splitlines():
-            nodename, state = out.split(" ", 2)
-            if state in ('alloc', 'comp'):
+            try:
+                nodename, state, features = out.split("|", 3)
+            except ValueError:
+                continue
+            if state in ('alloc', 'alloc*',
+                         'comp',  'comp*',
+                         'mix',   'mix*',
+                         'drng',  'drng*'):
                 nodestates[nodename] = 'busy'
-            elif state == 'idle':
-                nodestates[nodename] = 'idle'
+            elif state in ('idle', 'fail'):
+                nodestates[nodename] = state
             else:
                 nodestates[nodename] = 'down'
+            if features != "(null)":
+                nodefeatures[nodename] = features
 
         for n in nodelist:
             if n["slot_number"] and n["hostname"] and n["hostname"] in nodestates:
                 n["crunch_worker_state"] = nodestates[n["hostname"]]
             else:
                 n["crunch_worker_state"] = 'down'
+            n["slurm_node_features"] = nodefeatures.get(n["hostname"], "")
 
         return nodelist
 
@@ -52,12 +66,22 @@ class CloudNodeListMonitorActor(clientactor.RemotePollLoopActor):
     nodes, and sends it to subscribers.
     """
 
+    def __init__(self, client, timer_actor, server_calc, *args, **kwargs):
+        super(CloudNodeListMonitorActor, self).__init__(
+            client, timer_actor, *args, **kwargs)
+        self._calculator = server_calc
+
     def is_common_error(self, exception):
-        return self._client.is_cloud_exception(exception)
+        return isinstance(exception, config.CLOUD_ERRORS)
 
     def _item_key(self, node):
         return node.id
 
     def _send_request(self):
-        n = self._client.list_nodes()
-        return n
+        nodes = self._client.list_nodes()
+        for n in nodes:
+            # Replace the libcloud NodeSize object with compatible
+            # CloudSizeWrapper object which merges the size info reported from
+            # the cloud with size information from the configuration file.
+            n.size = self._calculator.find_size(n.extra['arvados_node_size'])
+        return nodes